[PPT] - Microarray Data Analysis of Adenocarcinoma Patients Survival Using PowerPoint Presentation

SLIDE 1

Wenting Zhou, Weichen Wu, Nathan Palmer, Emily Mower, Noah Daniels, Lenore Cowen, Anselm Blumer Tufts University http://camda.cs.tufts.edu

Microarray Data Analysis of Adenocarcinoma Patients’ Survival Using ADC and K-Medians Clustering

SLIDE 2

Overview

Goals Introduction Explanation of ADC and NSM Explanation of MVR, K-Medians, and

Hierarchical Clustering

Results Conclusions

SLIDE 3

Goals

Start with a classification of patients into

high-risk and low-risk clusters

Obtain a small subset of genes that still leads

to good clusters

These genes may be biologically significant One can use statistical or machine learning

techniques on the reduced set that would have led to overfitting on the original set

SLIDE 4

Introduction

We applied clustering and dimension-reduction

techniques to gene expression values and survival times of patients with lung adenocarcinomas

GENE AD10 AD2 AD3 AD5 AD6 AD7 AD8 L01 L02 L04 GABRA3 170 59.7 80 92.4 104 88 69.7 230 105 53.7 OMD 69.4 18.1 26 96.9 72.8 138.6 11.1 176 78.1 36.7 GS3686 250.7 146.8 150 177.8 228.7 115.5 177.8 511.3 233.9 393.6 SEMA3C 957.1 186.8 340.2 515.8 540.8 616.6 380.5 523.9 602.7 160.5 GML 25.4

7.7
16.3

18 26 9 21 32 24.3 27 MKNK1 471.2 309 225.7 296.6 264.1 371.9 291 664.2 471.6 407.3 OGG1

52
99

23.5 48.5

10

49.2

62.5
17.1

20

4.4

VRK1 42.8 57.9 69.4 60.4 56.4 37.2 99 295 78.1 94.2 VRK2 200.9 151.5 207.6 151.5 145.9 149.2 238.8 607.2 300.7 411 RES4-22 846.4 722.8 515.1 819.1 674.4 618.9 936.2 1388.1 732.1 959.1 SH3BP2 134.7 55.3 63.7 56.3 122.6 49.2 139.3 362.5 115.5 52 NULL 147 131.2 107 118.9 174 92 175.9 396.9 90 185.3 NULL

71.4
85.4
78.3
80.7
85.2
135.3

4.1 46

76.4
50.2

RES4-25 19.6

44

49.2 22.2

69.2

17 6.8 60 81 105 RNF4 953.2 552.1 609.4 708.2 582.7 768.1 1130.1 1062.6 1005.8 1561.9 PLAB 703.6 2068.7 447 2771.2 327.1 179 1427.8 460.4 3691.9 1583.4 ARNTL 22.2

22

30.8 75.5 32 57 28.2 47 34.8 34.3 CDH23 222.2 178.3 99 111.6 157.1 133.2 340.2 325 131.9 181.5 PCDHGB4 43.5 69 53.4 67.6 66.8 60 45.8 125 66.8 76.4 PCDHGA12

7
0.8

28.4 4.2 3

0.6

6.8 1 10.4 2.3 H4FM 95.5 75.1 68.5 57 35.5 54.5 55.1 152.6 71.1 88 GMFB 526.9 391.8 288.9 326.1 383.1 416.4 806.9 1286.3 669.6 437.3 AQP3 777.5 517.9 1053.2 4190.3 449.5 421.9 709.9 687 1194.1 413.8 KIAA0316 62.3 52 24.8 43.8 31 39 45.8 162.6 44 48.5 KIAA0317 149 328.6 199.4 172 288 321.4 238.8 314.7 201.8 298 KIAA0320 565.7 467.2 378 522.1 558.9 432.1 571.7 592.4 493.8 517.2 CLOCK 400.6 259.7 238.5 400 340.5 360.3 189.1 365.3 252.6 433.8 MADD 554.6 480.9 528.7 618.6 530 471.1 597.3 486.3 427 393.6 KIAA0367 68.5 65 16 108 32 98 95.8 195.1 52.8 15 KIAA0368 22.2 4 10.8 70.2 23.5 35.5 41 84.6 43 31 ARHGEF12 281.6 355.7 650.7 795.5 412.5 371.9 246.8 437 375.8 454.9 CTNND1 1018.2 1579.4 1254.4 1293.3 1220 1053.2 1098.5 738.6 703.6 3401.2 SCYA21 658.2 419.8 319.3 172 358.5 315.2 426.1 510.5 190.8 350.6

gene AD-043T2-A7-1 AD-111T2-A8-1 AD-114T1-A9-1 * AD-115T1-A12-1 * AD-118t1-A13-1 AD-119t3-A195-8 AD-120t1-A226-8 * AD-122t3-A197-8 interleukin 2

18.6

9.12

2.175
1.54
9.07
16.58
15.895
14.5

interleukin 10 10.54 9.12

2.21

21.75 3.08

20.09

10.88

10.48

interleukin 4 0.01 10.18

0.06

5.835

1.98
8.39

1.61 3.61 tumor necrosis factor receptor superfamily, member 6 19.44 29.29 6.32 23.815 17.26 4.49 23.845 12.67 J04423 E coli bioB gene biotin synthetase (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively)

16.98
4.68
1.775
24.785
10.09
18.92
21.98
17.52

J04423 E coli bioB gene biotin synthetase (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively)

27.5
1.5
16.53
12.89
15.15
20.09
29
20.54

J04423 E coli bioB gene biotin synthetase (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively)

1.6
3.62
3.61
4.485
18.19
8.39
3.865

0.59 J04423 E coli bioC protein (-5 and -3 represent transcript regions 5 prime and 3 prime respectively) 38.88 20.8 16.41 19.5 13.21 16.19 23.635 28.78 J04423 E coli bioC protein (-5 and -3 represent transcript regions 5 prime and 3 prime respectively)

29.12
13.18
17.97
21.445
13.13
38.82
19.01
22.55

J04423 E coli bioD gene dethiobiotin synthetase (-5 and -3 represent transcript regions 5 prime and 3 prime respectively)

42.87
35.47
57.02
47.205
39.47
56.38
65.195
68.78

J04423 E coli bioD gene dethiobiotin synthetase (-5 and -3 represent transcript regions 5 prime and 3 prime respectively) 121.62 50.53 59.36 46.995 53.71 68.85 71.025 78.18 X03453 Bacteriophage P1 cre recombinase protein (-5 and -3 represent transcript regions 5 prime and 3 prime respectively)

22.64
14.24
19.73
7.555
30.35
15.41
22.815
22.55

X03453 Bacteriophage P1 cre recombinase protein (-5 and -3 represent transcript regions 5 prime and 3 prime respectively) 2.44 10.18 2.99 12.885

3
4.87

0.965 4.62 J04423 E coli bioB gene biotin synthetase (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively) 51.04 86.63 29.485 112.72 74.96 19.71 93.535 54.99 J04423 E coli bioB gene biotin synthetase (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively) 14.59

5.74
4.765
35.865
1.98

0.98

30.79
35.62

J04423 E coli bioB gene biotin synthetase (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively)

97.84
43.96
65.625
61.04
79
56.38
97.25
111.96

J04423 E coli bioC protein (-5 and -3 represent transcript regions 5 prime and 3 prime respectively)

38.82
3.62
32.87
26.21
19.2
24.77
31.695
31.6

J04423 E coli bioC protein (-5 and -3 represent transcript regions 5 prime and 3 prime respectively)

7.27
5.74
11.285
6.535
11.1
35.31
7.655
25.56

J04423 E coli bioD gene dethiobiotin synthetase (-5 and -3 represent transcript regions 5 prime and 3 prime respectively)

34.78

10.18

12.12

18.265

10.09
4.87

19.03

5.45

J04423 E coli bioD gene dethiobiotin synthetase (-5 and -3 represent transcript regions 5 prime and 3 prime respectively) 34.02 13.37 6.805 20.2

8.06
16.58

8.025 39.87 X03453 Bacteriophage P1 cre recombinase protein (-5 and -3 represent transcript regions 5 prime and 3 prime respectively)

12.13

9.12

10.245
5.04
7.05
13.07
13.15
18.52

X03453 Bacteriophage P1 cre recombinase protein (-5 and -3 represent transcript regions 5 prime and 3 prime respectively)

60.66
9.99
22.565
26.475
46.57
58.73
46
52.71

U14573 Human Alu-Sq subfamily consensus sequence. 7322.58 5795.86 8056.02 6437.37 7254.32 6222 6715.07 6766.43 L38424 B subtilis dapB, jojF, jojG genes corresponding to nucleotides 1358-3197 of L38424 (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively) 4.06 20.8 2.285 12.87 1.06

3.7

11.67 5.63 L38424 B subtilis dapB, jojF, jojG genes corresponding to nucleotides 1358-3197 of L38424 (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively) 21.06 30.36 9.79 32.835 13.21 0.98 24.68 30.8 L38424 B subtilis dapB, jojF, jojG genes corresponding to nucleotides 1358-3197 of L38424 (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively)

15.36

3.81

4.295

3.38

6.03
9.56
0.745
5.45

X17013 B subtilis lys gene for diaminopimelate decarboxylase corresponding to nucleotides 350-1345 of X17013 (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively) 0.01 16.55 4.62 7.395

11.1
3.7

0.6 0.59 X17013 B subtilis lys gene for diaminopimelate decarboxylase corresponding to nucleotides 350-1345 of X17013 (-5, -M, -3 represent transcript regions 5 prime, Middle, and 3 prime respectively)

11.32
5.74
11.15
9.455
23.26
30.63
14.36
10.48

Harvard Data (n= 84) Michigan Data (n= 86) 12,600 genes 7129 genes

SLIDE 5

Overview

Goals Introduction Explanation of ADC and NSM Explanation of MVR, K-Medians, and

Hierarchical Clustering

Results Conclusions

SLIDE 6

ADC and NSM Overview

We use Approximate Distance Clustering

maps (Cowen, 1997) to project the data into

ne or two dimensions so we can use very

simple clustering techniques.

Then we use Nearest Shrunken Mean

(Tibshirani, 1999) to reduce the number of

genes used to predict the clusters.

We evaluate using leave-one-out

crossvalidation and log-rank tests

SLIDE 7

Approximate Distance Clustering (ADC, Cowen 1997)

Approximate Distance Clustering is a method

that reduces the dimensionality of the data.

This is done by calculating the distance from

each datapoint to a subset of the data, which is called a witness set.

A different witness set is used for each

desired dimension

A simple clustering technique is used on the

projected data

SLIDE 8

ADC map in one dimension

SLIDE 9

1-d ADC map with cutoff

SLIDE 10

General ADC Definition

Choose witness sets D1, D2, …, Dq to be

subsets of the data of sizes k1, k2, …, kq

The associated ADC map

f(D1, D2, …, Dq) : Rp Rq maps a datapoint x to (y1, y2, …, yq) where yi = min{ || xj – x || : xj ∈ Di} is

the distance to the closest point in Di

SLIDE 11

Criterion for a good clustering

Compute the Kaplan-Meier survival curves and the

p-value from the log-rank test, then choose the clustering that minimizes:

W = 4000*a + 5500*b + 450*(1-c)+ 50*d where

a= 1 if the size of the smaller group < n/8 and 0 otherwise b is the p-value c is the difference between the final survival rates of the

low-risk and high-risk groups

d is the high-risk group’s final survival rate

SLIDE 12

Nearest Shrunken Mean (NSM) Gene Reduction (Tibshirani,1999)

NSM eliminates genes with cluster means

close to the overall mean.

NSM shrinks the cluster means toward the

verall mean by an amount proportional to

the within-class standard deviations for each gene.

If the cluster means all reach the overall

mean, that gene can be eliminated.

SLIDE 13

Definition of NSM

This gene would be retained

Overall mean for gene i Cluster 1 mean Cluster 2 mean

(s0+si) m1 ? (s0+si) m2 ?

SLIDE 14

Definition of NSM

This gene would also be retained

Overall mean for gene i Cluster 1 mean Cluster 2 mean

(s0+si) m1 ? (s0+si) m2 ?

SLIDE 15

Definition of NSM

This gene would be eliminated

Overall mean for gene i Cluster 1 mean Cluster 2 mean

(s0+si) m1 ? (s0+si) m2 ?

SLIDE 16

Overview

Goals Introduction Explanation of ADC and NSM Explanation of MVR, K-Medians, and

Hierarchical Clustering

Results Conclusions

SLIDE 17

MVR and K-Medians Overview

We use naïve clustering by survival time

instead of ADC for the initial clusters

We use variance ratios instead of NSM We reduce genes further using hierarchical

clustering of expression profiles

We evaluate using K-medians and log-rank

tests

SLIDE 18

Method: Minimum Variance Ratio (MVR) Gene Reduction

The variance ratio is the sum of the

within-cluster variances divided by the total variance of expression values for that gene.

Genes with large variance ratios are

thought to contribute less to the cluster definitions and are eliminated.

SLIDE 19

Hierarchical Clustering of Genes

Different genes may have similar expression

profiles

Eliminating similar genes may lead to a

smaller set of genes that still leads to a good separation into high-risk and low-risk clusters

Hierarchically cluster the genes until the

desired number of clusters is obtained, then select one gene from each cluster

SLIDE 20

K-Medians Clustering

This unsupervised clustering method finds the

K datapoints that are the best cluster centers

In this paper we use K= 2 so it is possible to

find the optimal clustering by trying all possible pairs of points as cluster centers.

The quality of the clustering is calculated as

the total distance of data points to their cluster centers

SLIDE 21

Overview

Goals Introduction Explanation of ADC and NSM Explanation of MVR, K-Medians, and

Hierarchical Clustering

Results Conclusions

SLIDE 22

Experimental Results

The following tables give the results obtained when

using the W criterion to select the best ADC witnesses and cutoffs, then reducing the set of genes with NSM.

The p-values were obtained from leave-one-out

crossvalidation on the reduced set of genes.

The values for STCC were obtained by following the

same procedure but substituting clusters formed from the 50% or 60% highest risk patients for the ADC clusters.

SLIDE 23

Comparison of 1-d and 2-d ADC with STCC on Michigan data (n = 86)

p- value Lo w-ris k/hig h-ris k gro up si ze Nu mber

f ge

nes 1D 1K AD C 2D 1K AD C 50% STC C 60% STC C 1D 1K AD C 2D1K ADC 50% STC C 60 % STC C

7129 0.0028 0.0500 0.0086 0.0126 55/31 54/32 46/40 46/40 1000 0.0275 0.0009 0.0111 0.0158 59/27 60/26 45/41 43/43 500 0.0495 0.0048 0.0046 0.0089 52/34 57/29 47/39 45/41 200 0.0019 0.0033 0.0075 0.0056 58/28 58/28 47/39 48/38 100 0.0058 0.0194 0.0023 0.0048 57/29 55/31 49/37 46/40 50 0.0019 0.1442 0.0064 0.0048 58/28 42/44 50/36 47/39 40 0.0009 0.0268 0.0011 0.0048 58/28 44/42 50/36 47/39 30 0.0009 0.0356 0.0029 0.0067 58/28 43/43 51/35 46/40 20 0.0021 0.0189 0.0029 0.0090 57/29 42/44 51/35 46/40 10 0.0061 0.0618 0.0059 0.0049 56/30 37/49 50/36 47/39 5 0.0086 0.3559 0.0151 0.0024 58/28 41/45 49/37 49/47

SLIDE 24

Kaplan-Meier Curve (p= .0009)

Michigan 1-D ADC, 40 genes

0.2 0.4 0.6 0.8 1 1.2 20 40 60 80 100 120 Time (Month) Survival Rate low risk censored high risk censored

SLIDE 25

Comparison of 1-d and 2-d ADC with STCC on Harvard data (n = 84)

p- value Lo w-ris k/hig h-ris k gro up si ze Nu mber

f ge

nes 1D 1K AD C 2D 1K AD C 50 % STC C 60 % STC C 1D 1K AD C 2D 1K AD C 50 % STC C 60 % STC C

12600 0.0646 0.0046 0.1946 0.0741 25/59 24/60 39/45 41/43 1000 0.0124 0.0013 0.0381 0.0038 20/64 15/69 44/40 38/46 500 0.0023 0.0116 0.0021 0.0027 21/63 22/62 42/42 36/48 200 0.0121 0.0037 0.0007 0.0004 21/63 21/63 40/44 32/52 100 0.0201 0.0027 0.0213 0.0004 24/60 26/58 42/42 30/54 50 0.0332 0.0090 0.0120 0.0047 21/63 21/63 40/44 35/49 40 0.0332 0.0019 0.0120 0.0033 21/63 27/57 40/44 35/49 30 0.0898 0.0010 0.0065 0.0098 28/56 26/58 39/45 35/49 20 0.0448 0.0039 0.0083 0.0015 27/55 26/58 38/46 34/50 10 0.0424 0.0011 0.0034 0.0001 22/62 20/64 37/47 33/51 5 0.0321 0.0032 0.0053 0.0196 20/64 25/59 36/48 28/56

SLIDE 26

Kaplan-Meier Curve (p= .0332)

Harvard 1-D ADC, 40 genes

0.2 0.4 0.6 0.8 1 1.2 20 40 60 80 100 120 Time (Month) Survival Rate low risk censored high risk censored

SLIDE 27

Validating ADC Between Michigan and Harvard Data

We validated the 100 genes we obtained

from the Michigan data by finding the genes in the Harvard data that matched by gene symbol and using those to run leave-one-out crossvalidation on the Harvard data.

For the 1-dimensional ADC, we found 88

matching genes in the Harvard data and

btained a p-value of 0.0076 with cluster

sizes of 25 and 59.

SLIDE 28

Kaplan-Meier Curve (p= .0076)

Using 1-d ADC by using the Michigan based top 100 survival genes to identify a low- and high-risk group on Harvard data

Independent test: Harvard 1-D ADC, 88 matched genes 0.2 0.4 0.6 0.8 1 1.2 20 40 60 80 100 120 Time (Month) Survival Rate low risk censored high risk censored

SLIDE 29

Validating ADC Between Validating ADC Between Michigan and Harvard Data Michigan and Harvard Data

We validated the 100 genes we obtained

We validated the 100 genes we obtained from the Harvard data by finding the genes in from the Harvard data by finding the genes in the Michigan data that matched by gene the Michigan data that matched by gene symbol and using those to run leave-one-out symbol and using those to run leave-one-out crossvalidation crossvalidation on the Michigan data.

n the Michigan data.

For the 1-dimensional ADC, we found 70

For the 1-dimensional ADC, we found 70 matching genes in the Michigan data and matching genes in the Michigan data and

btained a p-value of 0.0495 with cluster
btained a p-value of 0.0495 with cluster

sizes of 22 and 64. sizes of 22 and 64.

SLIDE 30

Kaplan-Meier Curve (p= .0495) Kaplan-Meier Curve (p= .0495)

Using 1-D ADC by using the Harvard based top 100 survival Using 1-D ADC by using the Harvard based top 100 survival genes to identify a low- and high-risk group on Michigan data genes to identify a low- and high-risk group on Michigan data

Independent test: Michigan 1-D ADC, 70 matched genes

0.2 0.4 0.6 0.8 1 1.2 20 40 60 80 100 120 Time (Month) Survival Rate low risk censored high risk censored

SLIDE 31

Top survival-related genes Top survival-related genes

13 common Genes of Michigan top 100 genes and 13 common Genes of Michigan top 100 genes and Harvard top 100 genes using 1-D ADC Harvard top 100 genes using 1-D ADC

Symbol Name CD37 CD37 antigen CD74 CD74 antigen (invariant polypeptide of major histocompatibility complex, class II antigen- associated) GAPD glyceraldehyde -3-phosphate dehyd rogenase HE1 epididymal secretory protein (19.5kD) HLA-DMA major histocompatibility complex, class II, DM alpha HLA-DMB major histocompatibility complex, class II, DM beta HLA-DPB1 major histocompatibility complex, class II, DP beta 1 HLA-DQB1 major histocompatibility complex, class II, DQ beta 1 HLA-DRA major histocompatibility complex, class II, DR alpha HLA-DRB1 major histocompatibility complex, class II, DR beta 1 MIF macrophage migration inhibitory factor (glycosylation-inhibiting factor) PFN2 profilin 2 SFRS9 splicing factor, arginine/serine-rich 9

SLIDE 32

MVR and K-Medians results

We used Minimal Variance Ratio to select 200 genes

from the Michigan and Harvard data based on an initial 50-50 clustering according to survival times.

We then used hierarchical clustering to group these

genes into 40 clusters.

We selected one gene from each cluster and

performed a K-medians clustering of the patients into a high-risk and low-risk group using these 40 genes after normalizing their expression profiles so that the clusters wouldn’t be unduly influenced by genes with high mean expression values.

SLIDE 33

MVR and K-Medians results

On the Michigan data this gave a p-value of 0.00002

with cluster sizes of 36 and 50, while on the Harvard data the p-value was 0.0417 with cluster sizes of 47 and 37.

We used leave-one-out crossvalidation to verify this

whole procedure.

After clustering, the remaining patient was classified

as high-risk or low-risk according to which cluster had the smaller average distance to that patient.

For the Michigan data, this gave a p-value of 0.0219

and for the Harvard data the p-value was 0.0696.

SLIDE 34

Kaplan-Meier Curve (p= .00002)

Michigan 40 genes from 200

0.2 0.4 0.6 0.8 1 1.2 20 40 60 80 100 120 Time (Month) Survival Rate high risk censored low risk censored

SLIDE 35

Kaplan-Meier Curve (p= .0417)

Harvard 40 genes from 200

0.2 0.4 0.6 0.8 1 1.2 20 40 60 80 100 120 Time (Month) Survival Rate low risk censored high risk censored

SLIDE 36

Conclusion

Combinations of simple techniques yield

small sets of genes with high predictive power

Different techniques give different sets

f genes

ADC - NSM was often superior to MVR -

K-medians - Hierarchical Clustering, but the latter was surprisingly good

SLIDE 37